AI services in cloud based image management: What to consider

By teamnext Editorial Team

Using artificial intelligence to process large volumes of images is effective and has gained significant relevance in recent years. Within a short time, faces can be marked across thousands of photos to group the material by person. Face recognition is only one prominent special case.

Based on machine learning, it is also possible to identify other objects with similar characteristics, provided that suitable training has taken place. Today, AI services can recognize and classify many everyday objects in images. They then generate relevant metadata such as keywords or broader categories. Even repositories with millions of images can be indexed within a day using conventional computing resources.

Dachshund on leather sofa - symbol image Color recognition and auto-tagging by artificial intelligence - with the professional media management of teamnext
Automatic color and object recognition, symbol image.

Automatic color and object recognition, symbolic image.

Processes that require language processing can also be accelerated by AI. A related example is an article about ChatGPT in media management. In addition, AI technology is relevant for generating new images, for example through DALL-E or Stable Diffusion.

The industry largely relies on major providers

It is well known that internet companies and intelligence services use such technologies for their own purposes. It is also unsurprising that cutting edge technology is primarily developed in large corporations or well funded research institutions. Companies such as Microsoft or Google advance artificial intelligence not only for internal use. Licensing proprietary AI software to third parties is at least as important as a revenue driver.

Licensees are business customers who want to integrate AI algorithms into their products but want to avoid the cost and time required for in house development. This also applies to many providers of DAM systems. In Germany, AI in image management has been recognized and implemented as a major trend. However, and this is a problem, many solutions largely rely on major providers based in the United States, which are not subject to European Union data protection law.

Data sovereignty requires software developed in house

This data protection gap is rarely disclosed voluntarily by vendors. Many users also do not realize how strongly software increasingly depends on external components. The common logic is simple. There is no need to reinvent the wheel. Vendors often prefer open source components. Some innovations, however, are not available as open source and can only be used under license. In that context, this is referred to as proprietary software.

In practical terms, if large parts of a solution depend on services running on servers, often located in the United States, data sovereignty is limited. Face recognition is a good example. It involves sensitive data, namely biometric information related to facial symmetry, which should not fall into the wrong hands. If an image management system is intended to process personal data, data sovereignty must be addressed explicitly during vendor selection.

Ideally, a provider should contractually guarantee the following:

  • All software modules are developed in house and in Germany

  • Hosting exclusively within the European Union, preferably in Germany

  • Compliant storage of biometric similarity vectors

  • Continuous development, with a strong focus on data security

Only if these points are contractually defined is there legal certainty and a reasonable basis to assume that data will not leave the defined scope.

Stay flexible with a trainable AI solution

Flexibility is another key criterion when procuring AI based image management software. An AI platform is of limited value if it can recognize only everyday objects but not a specific product or an individual logo.

It is therefore important to choose software that enables training AI modules easily and with a high recognition rate.

Choose a transparent billing model

Pricing also needs attention. An innovative solution with proprietary AI services and strict data protection standards typically comes at a cost. Cloud based solutions are usually billed via a SaaS licensing model. The key is that the pricing model is transparent, including how AI functionality is charged.

Relevant questions include:

  • Are there additional costs for using AI modules

  • If yes, is pricing based on a flat fee or on usage

  • If usage based, is billing tied to documented AI actions, for example a tracked number of AI operations